The role of data mining technology in the identification of signals of possible adverse drug reactions: value and limitations

Abstract

Background: The emergence of large databases of adverse event (AE) data and the need to identify signals of new, unknown adverse effects of newly marketed drugs by regulators and pharmaceutical sponsors have coincided with the development of several methods of “data mining” for identifying new associations within all types of databases. Objective: This paper provides a broad overview of the data mining methods being used in many fields to consider applications for identifying new AEs in spontaneous AE databases and other medical data sources (eg, clinical trials and claims data). Methods: Literature was obtained through a MEDLINE search of the medical literature and a broader search of the medical informatics literature on the Internet. Results: Data mining methods have emerged to define associations in many types of databases. Specific methods include artificial neural networks, Bayesian probability approaches, genetic algorithms, decision trees, nearest neighbor methods, rule induction, and new data visualization techniques. Application of selected methods is now under way at the US Food and Drug Administration and the World Health Organization Centre for Drug Monitoring, as well as in some commercial organizations. Whether such methods enhance the usual AE signal identification process remains controversial. The application of data mining to coherent population-based clinical trial and epidemiological clinical data sets will likely enhance the AE field. Conclusion: Data mining methods show promise for the identification of new, unknown signals of AEs, especially in defined populations.